Abstract
The Mizar system is one of the pioneering systems aimed at supporting mathematical proof development on a computer that have laid the groundwork for and eventually have evolved into modern interactive proof assistants. We claim that an important milestone in the development of these systems was the creation of organized libraries accumulating all previously available formalized knowledge in such a way that new works could effectively reuse all previously collected notions. In the case of Mizar, the turning point of its development was the decision to start building the Mizar Mathematical Library as a centrallymanaged knowledge base maintained together with the formalization language and the verification system. In this paper we show the process of forming this library, the evolution of its design principles, and also present some data showing its current use with the modern version of the Mizar proof checker, but also as a rich corpus of semantically linked mathematical data in various areas including webbased and natural language proof presentation, maths education, and machine learning based automated theorem proving.
Introduction
Around 1970s, the advances in computer technology and its popularization together with the proliferation of more userfriendly programming languages allowed the mathematical community to initiate several seminal projects like de Bruijn’s Automath [20], Milner’s LCF [58] or Glushkov’s Evidence Algorithm [54]. The Mizar project [9] started in 1973 under the leadership of Andrzej Trybulec, first at the Płock Scientific Society and since 1976 at the University of Białystok (formerly the University of Warsaw, Białystok Branch), Poland. From the very beginning Trybulec postulated a language and a computer system for recording mathematical papers in such a way that [35, 56]: (a) the papers could be stored in a computer and later, at least partially, translated into natural languages, (b) the papers would be formal and concise, (c) it would form a basis for construction of an automated information system for mathematics, (d) it would facilitate detection of errors, verification of references, elimination of repeated theorems, etc., (e) it would open a way to machine assisted education of the art of proving theorems, (f) it would enable automated generation of input into typesetting systems. The initial ideas are still valid and with time and a growing support from more researchers involved in the project the current development can be geared towards more ambitious goals, offering more intelligent proof checking methods and better support for the users. A crucial factor that helped to establish Mizar’s position among leading proof assistants, stand the test of time and consequently be developed in the direction of achieving these goals, was the realization of the fact, that largescale formalizations require developing methods of efficient accumulating, maintaining and reuse of previously generated mathematical content. This approach is today taken by developers of dedicated formal libraries e.g. the Isabelle based Archive of Formal Proofs [14], as well as all large formalization projects, both in mathematics (like the Hales’s Flyspeck project [36], G. Gonthier’s formalization of the Feit–Thompson theorem [25]) as well as in formal computer science (e.g. NASA PVS Library [18], seL4: Formal Verification of an OperatingSystem Kernel [47], etc.).
The Beginnings of Mizar Mathematical Library
From the beginning of the Mizar project, the encoding of mathematical proofs was conducted in a dedicated formal language – the Mizar language. The formalization scripts were stored in plain text files, called articles, processed independently and with little connection to one another. The 1981 version of Mizar2 introduced the environment part of an article, at that time containing statements that were used as axioms and checked only syntactically, i.e., without requiring their justification. Later versions (Mizar3 and Mizar4 from the period 1982–1988) divided the processing of Mizar files into multiple passes with filebased communication, such as scanning, parsing, type and naturaldeduction analysis, and justification checking. The use of special vocabulary files for symbols together with infix, prefix, postfix notation and their combinations resulted in greater closeness to mathematical texts.
In 1986, Mizar4 was ported to the IBM PC platform running MSDOS and later became PCMizar in 1988. In the years 1987–1991 the Mizar system and language played an important role in a Polish state research grant programme of the Polish Ministry of Science and Higher Education RPBP III.24 “Logical systems and algorithms for computerized checking of proof correctness”.
It was this project that significantly boosted the development of Mizar articles by numerous authors (about 100 researchers and students from a dozen of scientific institutions from all over Poland), whose efforts resulted in almost 250 Mizar articles. Some of the articles were quite advanced. E.g. Trybulec formalized a 1970 paper by Borsuk “On the Homotopy Type of Some Decomposition Spaces” [17]^{Footnote 1} and his own original result that the algebra of normal forms is a Heyting algebra [83, 84]^{Footnote 2}. Some of the articles resulted from an experimental Mizaraided topology course where students were supposed to formalize theorems previously introduced informally during lectures. The methodology of this course described by S. Czuba and A. Zalewska in the RPBP III.24 report from 1988 was the following:
Each student had to solve several tasks from a given group of tasks. Initially, it was assumed that the students can reuse in their environment only the theorems and facts previously proved by others. But soon it turned out that this restriction cannot be applied to the simple facts from set theory (very often used in proving topological statements) obvious for students, while at the same time sometimes demanding tedious proofs. Therefore, each task contained in the description of its environment statements of theorems from set theory that were needed to solve a given task.
For each task group a special archive file was created, which was subsequently updated by adding new theorems together with their correct proofs. The care over the file archive was entrusted to one of the students, whose task was to take care of, among other things, the proper order of statements being added, as well as the integrity of operations and relations introduced by the students.^{Footnote 3}
Trying to maintain all these works naturally led to issues concerning the reusability of formalizations. In the RPBP III.24 report from 1988 E. Woronowicz wrote:
In previous versions of the Mizar system, including Mizar4, the text processed by the Mizar processor consists of two main parts:
declaration of the environment with definitions of basic concepts (modes, constants, functions) and statements of theorems playing the role of axioms for the developed theory,
the proper text, which consists of statements of new assertions and their proofs.
Such a text structure, mainly the fact that the user declares the environment, results in his full responsibility for the theory being developed (in the environment there may be contradictory items). Also the preparation time of Mizar articles is lengthened by the fact that when trying to correct errors e.g. in the proof of a statement, the processor processes the correct text of the environment each time. This is not a good solution neither in terms of methodology, nor technology.^{Footnote 4}
The first solution to address the reusability issue and avoid duplication of work was the implementation of a librarian utility which helped to include all the notions from previously processed files into a new article without the necessity to copy the whole environment by incorporating the contents of multiple files generated by several passes of the proof checker. This approach, however, turned out to be inefficient for bigger formalizations, which subsequently gave rise to developing methods of exporting only selected items from an article supported by corresponding changes in the Mizar language to facilitate both exporting and importing required notions. Later these concepts evolved into current forms used for importing theorems, definitions and other items from various articles. The functions of the librarian utility were superseded by extractor and accommodator, for extracting semantic information from an article into a database and including this information into a new formalization, respectively. From that time comes the conventional distinction between the library of Mizar articles (as a collection of userwritten files in the Mizar language) and the database consisting of the exported semantic information stored in dedicated machinereadable and optimized data formats. It was intended to avoid declaring adhoc axiomatics required for proving facts in each article, to help establish “a minimal axiomatization” for a particular article, to reduce the risk of introducing contradictory axioms and repetitions of already proved facts. January 1st, 1989 symbolically marks the date when the current Mizar library—Mizar Mathematical Library (MML) was born as the implementation of that proposal. The (still ongoing) process of building one common framework for verifying various branches of mathematics gave rise to a number of subsequent fundamental questions, e.g.:

should various axiomatizations be allowed or not (and if only one, then which one should be chosen);

how to build the knowledge database and what information should be stored in it;

how to organize, manage and ensure the integrity of the database.
In the following sections, we present how these issues have been addressed on the way to developing the current MML.^{Footnote 5}
Axiomatics
Owing to A. Trybulec’s mathematical background deeply rooted in the Polish school of mathematics, the library was from the beginning based on set theory. However, in the first years of the MML’s existence, there were several experiments with the foundations of the library. Namely, there were attempts to build the library based on set theory with classes (Morse–Kelley) and without them (Zermelo–Frænkel), as well as with the axiom of choice, or without it. Eventually, under the influence of formalizations in category theory, A. Trybulec decided on selecting the Tarski–Grothendieck set theory which is basically the Zermelo–Frænkel set theory with the usual extensionality, pair, union, regularity and replacement axioms, augmented with Tarski’s axiom of existence of arbitrarily large, strongly inaccessible cardinals [80] of the form:
For every set N there exists a system M of sets which satisfies the following conditions:
 (i)
\(N\in M\);
 (ii)
if \(X\in M\) and \(Y \subseteq X\), then \(Y\in M\);
 (iii)
if \(X\in M\) and Z is the system of all subsets of X, then \(Z \in M\);
 (iv)
if \(X \subseteq M\) and X and M do not have the same potency, then \(X\in M.\)
Technically, the initial axiomatization was introduced in two special axiomatic files: HIDDEN and TARSKI. The former contained a selection of primitive notions built into the checker e.g. the types Any and set, equality and membership relations, but also types Element of, Subset of, and the powerset operation bool. The latter presented only the statements of TG axioms. The idea was to keep the proof checking system independent from any particular set theory, and so the type Any was introduced to represent any arbitrary object, not necessarily being a proper set. This, in principle, enabled developing other libraries with different axiomatizations. However, the library developed in terms of TG contained an additional axiom that the type Any is also of type set. Since the interest in developing alternative libraries was rather small, later the type Any was completely removed from the axiomatic files in favor of using the type set alone. Interestingly, in Mizar Ver. 8.1.01 from 2012, to filter out some futile definitional expansions automatically generated by a newly implemented mechanism in the Mizar checker [52], the most general root type was restored into the HIDDEN file, but under a new name: object.
The current form of the TARSKI file representing the set theory axioms is the result of library reorganization performed in connection with exploring finegrained dependencies in the library [6] that took place in 2013. The original TARSKI file was split into TARSKI_0, TARSKI_A and TARSKI. The TARSKI_0 file contains only ZF axioms, TARSKI_A represents the Tarski’s axiom alone, and the TARSKI file contains more userfriendly formulations of the axioms proved as consequences of the raw axioms from the other two files, e.g. using a defined notion of an ordered set rather than the axiom that such a pair exists.
It should also be noted that the original axiomatics was extended by a selection of properties of some notions commonly used in various formalizations that were built into the system to improve its usability. The extra axiomatic file AXIOMS contained the definitional axioms of the following concepts: element, subset, Cartesian product, domain (non empty set), subdomain (non empty subset of a domain), set domain (domain consisting of sets), as well as the axioms of strong arithmetic of real numbers [76]. The process of restructuring this part of the library had two major steps: first the properties of settheoretic notions were consequently proved as consequences of basic axioms and moved to ordinary Mizar articles (i.e. ZFMISC_1, SUBSET_1), while the full stepbystep construction of real numbers was consequently defined in the ARYTM* series of articles in the years 1995–1998. The current form of the arithmetic in the MML was shaped around the year 2003 in the course of developing a series of encyclopedic articles XCMPLX* and XREAL* extracted from the library in order to simplify the browsing for selected useful properties of real, complex, and extended real numbers. In consequence, the sets of real and complex numbers were defined together with proving all their usual properties without resolving to any extra axioms, so the original article AXIOMS was removed.
On the other hand, the Mizar system was enhanced by introducing special automation of selected commonly used notions, as so called requirements [59]. In particular this concerned the arithmetic of complex numbers, so that the users could decide whether they want the arithmetic facts to be obvious for the Mizar checker [63, 68]. At the same time, the automatically obvious facts found their justification in corresponding ordinary Mizar articles to enable their crossverification and allow switching off the automation in special contexts, e.g. for educational purposes or for experimenting with axiomatic systems [6].
The separation and consequent use of the minimal axiomatization was an important landmark and since then has been the main approach taken in developing the current MML. This allows to shift the focus in extending the proof checking mechanisms from implementing hardcoded system procedures towards language extensions like registrations [62], properties [49, 60], or reductions [51].
MML Information Storage
The way in which the information is stored in the MML is a key property of the library. The basic library import/export unit is an article, resembling the conventional publication practice of mathematical papers. Initially, the article’s authors decide which definitions and statements they would like to export and make available to other formalizations, and which should be considered local only. In the process of peer review coordinated by the Library Committee^{Footnote 6} [30], all exportable items are finally selected and extracted to form the, so called, abstract, which contains the bare statements of definitions, theorems, schemes and registrations stripped of all their corresponding proof parts. This public interface information is subsequently stored in the database in dedicated XMLbased internal data formats. For any definition, theorem, scheme, or registration, the database contains the corresponding formula represented using constructors ^{Footnote 7}. Different kinds of constructors are used to refer to different classes of defined notions, see Table 1 for the use of constructors and their notations in the current MML.
The information stored in the database which corresponds to a definition, similarly holds a formula (the definiens), but also the definition’s notation which consists of its format and type information of its arguments, as well as the result type for functor definitions and the mother type for mode definitions. The format of a given definition contains the corresponding symbol (string of characters) together with the information on the arity and position of visible arguments in the text representation. The symbols grouped according to constructor kinds, possibly with additional priority declarations, are stored in, so called, vocabularies. Table 2 shows the number of symbols in each category.
The separation of vocabularies from the articles that first introduce given symbols facilitates the use of symbols for multiple, sometimes completely unrelated, notations. Table 3 presents the number of notations for most popular symbols.
The same format can also be used for arguments with different types, most importantly in the case of the redefinitions of previously defined notions with more specific type restrictions, see Table 4.
It should also be noted that selected definitions declare special properties of the defined notions to enable further automation in the proof checking software. The information stored in the current database with respect to these properties of predicates, functors and modes is shown in Table 5.
Apart from the theorems, schemes and definitions, the database stores also additional information extracted from the abstracts. That includes presentation related data of new notations (defining synonyms or antonyms for previously defined objects), but also automation, i.e. registrations^{Footnote 8}, as well as term reductions and identifications. Especially the registrations are heavily used in Mizar, see Table 6 (term reductions and identifications are also syntactically represented as registrations).
Since the formalization library imitates to some extent the cumulation of informal mathematical papers, it is interesting to measure the ratio of the number of proved theorems to the number of introduced definitions (cf. [14]). Taking into account only Mizar statements syntactically represented as theorems and schemes, one obtains a ratio of 4.95, see Table 7. However, various forms of Mizar registrations and properties also correspond to theorem statements in informal mathematics. If we also consider them as theorems, the ratio grows to 6.15, see Tables 5 and 6.
Importing all the aforementioned information from the database to the environment of new articles is done with the help of a set of designated directives in the Mizar language [34]. A first group of directives enables formulating and disambiguating formal Mizar texts considering overloaded notations i.e. vocabularies, constructors, notations, and partially also registrations and requirements (because of the overloading, the order of used notations is important [66]). Another group of directives concerns encoding proofs, i.e. specifying their skeletons (definitions) and conducting reasoning steps to justify a given goal (theorems and schemes). The rest of import directives influence the way in which the Mizar proof checker uses its built in automation procedures that offer shorter justifications of selected proof steps. This includes the directives requirements, registrations (also importing term identifications and reductions), expansions, and equalities (that automatically provide references to the definitions of used atomic formulas and terms, respectively) [26].
Various independently processed import directives in the Mizar language allow controlling the amount of imported information and accessing only the necessary items from the database. Thanks to this approach, processing information from the database should be similarly timeconsuming, despite the level of complexity of the formalization environment.
Organization and Management
The development of the MML has been driven by several main factors. Initially, most formalization attempts were targeted at covering background knowledge in chosen fields of mathematics. When the library was advanced enough, it was also possible to start works directed at formalizing individual theorems with nontrivial proofs. Moreover, several projects aimed to prove whole papers and monographs were carried out.
The first 3 years of the MML development were dominated by the first approach. The articles developed at that time covered mainly sets, relations, functions, and also geometry and general topology. However, there were also a few articles concerning category theory, functional analysis or quantum theory. Table 8 presents the number of articles (NOA) from that period according to the MSC classification scheme.
It should be noted that the formalizations from these initial years were mostly done independently with no actual connection between formalized theories (e.g. the hierarchies of functions and relations were disjoint, various geometries were constructed in different formal frameworks, groups, rings, fields, and vector spaces were all distinct, without any intrinsic hierarchy). Collecting as many articles as possible was the main priority then. It was the development of the library that eventually allowed to identify common works and emphasize on the quality of the formalizations and the organization of the collected library.^{Footnote 9}
A similar classification of the current MML content shows a certain trend, i.e. topology and real functions are still among the best developed domains (see Table 9). Notably, the field of mathematical applications in computer science, with a significant number of articles in the current MML, was hardly represented (there was just one article formalizing the basic concepts of Petri nets).
The largest area covered in the MML (in terms of the number of articles) are mathematical logic and foundations (MSC #03)—the repository contains the model of the Mizar language itself (the construction of the firstorder language, propositional tautologies and satisfiability, Gödel completeness theorem), but also the formalization of various systems of nonclassical propositional logic, including linear temporal time or Grzegorczyk’s logics. Alongside with the building blocks of Tarski–Grothendieck set theory (descriptive and combinatorial set theory and large cardinals), other systems based on nonstandard membership relation, such as fuzzy and rough sets [33], are also widely represented. The theory of ordered algebraic structures (MSC #06) follows closely the seminal handbook of G. Grätzer (for lattices and orders) and [23] (for domains). MSC #26 devoted to real functions is one of the most fundamental parts of the MML in terms of knowledge reuse, extending in a straightforward way classical set theory—essentially it covers the standard undergraduate course in mathematical analysis. We can point out that general algebraic systems [31] seem to be underrepresented—summing up the number of files from underlying AMS MSC sections (08, 12, 13, 15, 16, 20), we obtain 186 articles formalizing the classical S. Lang’s course of algebra.
Among the formalizations of particular theorems that stimulated the development of the library we can name an early article formalizing a fundamental theorem of functional analysis—the Hahn–Banach theorem [70] submitted to the MML in 1993. Formalizing the fundamental theorem of algebra [57] completed in 2000 served as an example of a parallel development in several different proof checking systems including HOL Light (2001) and Coq (2002). A collaborative effort to formalize an elementary proof of the Jordan curve theorem (JCT) [50] is notable for developing a vast number of facts concerning the twodimensional real space and properties of special sequences.
Along with the appearance of “The Hundred Greatest Theorems” list published by Paul and Jack Abad [1] and later expanded by Freek Wiedijk^{Footnote 10} the library gained another important source from which the theorems to be formalized have been selected by various Mizar authors. At the moment of writing this paper, the Mizar system, with a total of 65 verified theorems from the list, is placed at the third position among all systems involved.
As far as developing formalizations of more comprehensive pieces of mathematics is concerned, the most notable example was the project of translating A Compendium of Continuous Lattices (CCL) [23] to the Mizar language. It was considered as an important challenge in the spirit of the QED Manifesto [81]. In the introduction to its second edition, Continuous Lattices and Domains [24], the authors wrote:
This is also the place to report on an activity of the Mizar project group located primarily at the University of Bialystok, Poland, the University of Alberta, Edmonton, Canada, and the Shinshu University, Nagano, Japan. It is the aim of the Mizar project to codify mathematical knowledge in a database. The codification means the formalization of concepts and proofs mechanically checked for logical correctness. The Mizar language is a formal language derived from the mathematical vernacular. The principal idea was to design a language that is readable by mathematicians, and simultaneously, is sufficiently rigorous to enable processing and verifying by computer software.
Our monograph A Compendium of Continuous Lattices was chosen by the Mizar group for testing their system. Since 1995, the Compendium has been translated piece by piece into the language Mizar. As of August 2002, sixteen authors have worked on this specific project; they have produced fiftyseven Mizar articles.
That project stimulated an extensive development of lattice theory, both in algebraic and topological sense, together with relevant category theory results (compare the position of lattice theory in the list in Tables 8 and 9). On the other hand, a formalization of this size, carried out by an international collaborating group of developers, greatly influenced the development of the Mizar system [13]. The system had to efficiently cope with importing multiple theories, the users had to learn how to routinely work with and share a local database, some mechanisms were developed to deal with various representations of related objects in different formalisms (e.g. lattices as algebraic structures and as ordered sets) [28, 32]. Moreover, these developments created the possibility of formalizing selected papers from current research frontier in that field. For example, the welldeveloped theory of ordered sets served as a basis for translating a paper on betterquasiordering countable seriesparallel orders (cf. [82] and [74]).
The initial articles in the MML were produced by Polish researchers and students associated around the Mizar developers. The existence of a growing and centrally maintained library allowed a number of researchers from many research groups to collaborate. This in turn, allowed to build diverse parts of the library corresponding to the area of expertise of many people involved in the project. Table 10 shows that 6 out of 10 most productive MML authors came from countries other than Poland. Overall, the current MML contains contributions developed by authors from 18 countries, see Table 11.
The history of submissions presented in Fig. 1 shows an almost linear cumulative number of articles. At the same time one can observe a significantly higher number of formalizations submitted in years 1990, 1999 and 2004. The year 1990 was obviously connected with the intensive work on creating the MML, as well as the fact that a number of formalizations developed with earlier versions of Mizar were converted to the new version used for building the library. In 1999 there were many research groups involved in largescale collaborative formalizations, including the previously mentioned JCT and CCL Polish–Japanese–Canadian projects. The year 2004 was marked by an intensive development of functional analysis and calculus, mostly by Japanese and Chinese groups.
The interplay of formalizations developed by Mizar users from various groups is seen when analyzing the theorem dependence tree of Mizar articles based on the quantitative information transfer, i.e. the relation induced by the use of the by keyword for straightforward justification within proofs.
In this relation, an article A is an ancestor of an article B if B refers to theorems in A, that is A transfers information to B. The direct ancestor A of an article B is the article which transfers the largest quantity of information into B.
The amount of information that an article A transfers to an article B is calculated as the sum of information transferred by all theorems from A which are referred to in B.
Formally speaking, let T be a theorem from A that is referred to in B. The amount of information, I, carried into B by T is calculated using the Shannon formula:
where

a is the number of all references to T in B,

n is the number of all references to T in all articles at the time when B was the last article accepted into the MML,

N is the number of all references to theorems in the MML at the given time.
In the current MML, the total number of references is 642,978. The article with the maximal number of children (102) is FINSEQ_1 [10] which provides basic properties of finite sequences. The maximal level of the theorem dependence tree is 18, given by the following article chain including formalizations going from the axioms, ordinal numbers, the arithmetic of complex numbers, through Euclidean geometry to planimetry, developed by researchers from Poland, Canada, Japan, China, Italy and Belgium:

18.
EUCLID11 “Morley’s Trisector Theorem” by Roland Coghetto

17.
EUCLID10 “Some Facts about Trigonometry and Euclidean Geometry” by Roland Coghetto

16.
EUCLID_6 “Heron’s Formula and Ptolemy’s Theorem” by Marco Riccardi

15.
COMPLEX2 “Inner Products and Angles of Complex Numbers” by Wenpai Chang, Yatsuka Nakamura and Piotr Rudnicki

14.
COMPLEX1 “The Complex Numbers” by Czesław Byliński

13.
SQUARE_1 “Some Properties of Real Numbers. Operations: min, max, square, and square root” by Andrzej Trybulec and Czesław Byliński

12.
XREAL_1 “Real Numbers—Basic Theorems” by Library Committee

11.
XCMPLX_1 “Complex Numbers—Basic Theorems” by Library Committee

10.
XCMPLX_0 “Complex Numbers—Basic Definitions” by Library Committee

9.
ARYTM_0 “Introduction to Arithmetics” by Andrzej Trybulec

8.
ARYTM_1 “Non Negative Real Numbers. Part II” by Andrzej Trybulec

7.
ARYTM_2 “Non Negative Real Numbers. Part I” by Andrzej Trybulec

6.
ARYTM_3 “Arithmetic of Non Negative Rational Numbers” by Grzegorz Bancerek

5.
ORDINAL3 “Ordinal Arithmetics” by Grzegorz Bancerek

4.
ORDINAL2 “Sequences of Ordinal Numbers. Beginnings of Ordinal Arithmetics” by Grzegorz Bancerek

3.
ORDINAL1 “The Ordinal Numbers. Transfinite Induction and Defining by Transfinite Induction” by Grzegorz Bancerek

2.
TARSKI “Tarski Grothendieck Set Theory” by Andrzej Trybulec

1.
TARSKI_0 “Axioms of Tarski Grothendieck Set Theory” by Andrzej Trybulec
Figure 2 shows the distribution of maximal dependence for all MML articles, i.e. the maximal lengths of chains of dependent articles rooted in a given article.
From its beginnings, the MML was considered an experiment of practical formal modeling of mathematics in order to get a possibly wide group of collaborators. Hence the heterogeneous character of submissions was always taken into account, even if initially most of the authors were from one research group. When the number of collected texts allowed for more statistical methods of information management and the analysis of formalized knowledge, and when the complexity of encoded mathematics went higher and higher, the duties of the Library Committee evolved from a very liberal policy to accept virtually all submissions from the developers “as is”, through the stage of enhancing them in order to get possibly high level of uniformity, into the direction of gradual, continuous, and intensive mechanism of changes of the MML, called revisions. One may consider at least four different kinds of revisions:

an authored revision—consists of small changes in some articles in the library when the author of a new submission notices a possible generalization of already existing theorems or definitions. For such a task, usually it is necessary to improve some older articles that depend on the change. As a rule, however, a rather small part of the library is affected.

an automatic revision—takes place frequently whenever either a new revision software is developed (e.g. software for checking equivalence of theorems, which enables removing one or two equivalent theorems), or the Mizar verifier is strengthened and existing revision programs can use it to simplify articles [64, 69] or utilize newly implemented language features [53].

prettyprinting—if changes touch only the parts which are not exported to the library; when newly designed mechanisms allow shorter proofs [71,72,73].

a reorganization of the library—it involves changing the order of article processing used for (re)creating the Mizar database—the ordering is stored in a special file mml.lar distributed with each MML version.
In 2001, a significant reorganization was implemented [78]. Its purpose was to group together all the articles which did not depend on the notion of a structure and place their identifiers at the beginning of the mml.lar list. Two MML parts were respectively named as:

concrete, which does not use the notion of structure (set theory, relations, functions, arithmetic and so on);

abstract, i.e. the article STRUCT_0 and its descendants, all of them directly or indirectly using Mizar structures.
Obviously, these two parts were not meant to be completely independent—the concrete part could be used in the abstract one, but not vice versa.
Apart from that, it turned out to be convenient to also separate a part of the articles in which Random Access Turing Machines were modeled (named “SCM part”). Isolating these articles and placing them at the end of the mml.lar list enabled frequent revisions avoiding the need to constantly update the rest of the library.
Another part of the library was later distinguished as “Addenda”. Initially it comprised technical articles created in order to prove the correctness of the formal construction of real numbers, previously available as axioms only. This aimed at a gradual reduction of the axiomatic foundations of the Mizar library. Later, this part was extended with auxiliary notions and generalized parts extracted from various articles to enable better reuse of facts within the library. We can mention here the introduction of two big hierarchies of Mizar structures defined retrospectively in a more general way to gain better representation of ordinary mathematical practice [27]: introducing common formal frameworks for both saved 38,649 out of total 447,511 lines of code (LOC), i.e. 8.63% (counting only first 3 years of the MML) and resulted in 24 removed articles (Fig. 3).
A similar approach was taken when the Encyclopedia of Mathematics in Mizar, “EMM”, was formed as another distinctive part of the library in years 2002–2012. Its fourteen articles (with MML identifiers starting with a capital ‘X’) were extracted in order to simplify the browsing for selected, most commonly used notions and their properties (e.g. of real, complex, and extended real numbers, boolean properties of sets, families of subsets, ordered tuples, etc.).
In order to facilitate the process of enhancing the quality of the repository, a reviewing process for submissions to the MML was introduced in 2006. Adopting the commonly used scheme for ordinary mathematical journals accept/revise/reject introduced additional D (for Delay) grade for suggested MML revision. Consequently, the motives for revisions can be for example:

keeping the repository as small as possible,

preserving a clear organization of the repository in order to attract authors,

establishing “elegant” mathematics, e.g. using short definitions (without unnecessary properties) or better proofs.
As one of the the most important MML tools supporting a reviewer’s work, we can point out the MML Query [11] service. It proved its usefulness when subsequent EMM items were created. Also researchers, when writing their Mizar articles, can find it useful, but usually, typical author does not care too much if his lemma is already present in the library. Actually, searching for such auxiliary fact can take much more time than just proving it—this results in many repetitions in the library. This is the area where another tools can be useful: J. Urban’s prototype of a hammer [86] for the MML, called Mizar Proof Advisor, was primarily developed to serve as an assistant for authoring Mizar articles. The fast MoMM (Most of Mizar Matches) tool for fetching matching theorems [87], hence existing duplications can be detected and deleted from the MML [29].
Another popular software, MML CVS—the usual concurrent version system for the MML was active for quite some time, but then was postponed, because the changes were too cryptic for the reader due to the lack of proper marking of items. Actually, one of the most general problems is that there are no absolute names for MML items and the changes are sometimes too massive to find out what really matters; hence the usefulness of the tools of this type is very limited.
Usually, the revision process via automatic generalization of notions improves the MML. There are some problematic issues, however, which suggest that a careful supervision of human reviewers is definitely needed. For example, if we define an even integer number as follows [77]:
then, quite naturally, we can call odd all integers which are not even. As long as the assumption of a variable i to be integer is not needed in the formulation of the above definition, it can be marked for removal by the library software (it can be a real number, at least). But then, the number \(\pi \) can be shown to be odd as it is not even (although it is not an integer, hence such a classification is void). Any automation of the process of dropping assumption about the types of used loci in the definition of attributes, however useful from a general point of view, could be dangerous.
As a rule, building an extensive encyclopedia of knowledge needs some investment; on the one hand, it can be considered by purely financial means as “information wants to be free, people want to be paid” [2], and it can be observed that projects like the aforementioned formalization of CCL or JCT always go in hand with a significant expansion of the library.
Other Applications of the MML
The size and the diversity of the MML makes it not only an indispensable standard library for Mizar users, but also an important resource for various projects related to processing mathematical knowledge.
Among the main activities based on the content of the MML, we can mention the development of representations of formal mathematics in human readable LaTeX form (e.g. the Formalized Mathematics journal [12]), XMLbased semanticallylinked web pages [85, 90], more semantic variants of the Mizar language (e.g. WSMizar [61]), or general semanticallyrich formats like OMDoc [37]. The library’s content available under an open source license [4] was also used as a test bed for developing formal wikis [3]. Moreover, it is worthwhile that the development of first MMLbased semanticallylinked presentations of mathematical content in the form of the online Journal of Formalized Mathematics ^{Footnote 11} active in years 1995–2004 predated the advent of Wikipedia and other online maths journals and services popular today.
As already mentioned in Sect. 2, the development of the Mizar library since its beginnings has been connected with various educational projects involving earlier versions of Mizar. More recently, the organization and scientific content of several courses for mathematics and computer science students which used the modern Mizar versions 7 and 8 have been reported by Retel and Zalewska [75], Borak and Zalewska [16], as well as Naumowicz [65]. Over the last decade, a number of Mizar tutorials at conferences^{Footnote 12} and at summer schools^{Footnote 13} have been addressed to students and young researchers. The community of Mizar users has benefited from creating a number of introductory manuals^{Footnote 14} and a technical reference manual [34]. Some directions in the development of the Mizar proof checking system have also been targeted at beginner users, e.g. the works on making the MML more easily accessible using dedicated auxiliary tools [66] and simplified environment building [67].
Apart from various forms of presenting mathematics, the creation and development of the MML into a large corpus of formalized knowledge enabled experiments with mining the dependencies in formal mathematics [5], as well as training automated theorem provers [88]. The MML served as a basis for preparing largetheory problems to be solved by provers during the competitions of the CADE ATP System Competition [79] and provided a sufficiently big data for machine learning oriented ATP training [21, 40].
Among the most important projects related to the interplay of ATP and the MML we can mention here the J. Urban’s Mizar Proof Advisor, one of the first hammerstyle systems i.e. giving the authors of formal proofs a semiintelligent brute force tool that can take advantage of very large lemma libraries [15]. Thanks to these works Mizar gained ways of crossverification for its proof checking system by using external provers [89]. This technology was later adopted in the MizAR system which now provides an online automated reasoning service for Mizar users employing a large suite of AI/ATP methods trained over the Mizar library. Reportedly, the system is able to automatically prove 40% of the theorems from the whole MML [45].
The MML is of great value for various machine learning experiments not only for its size, but also for its inherent structure. The application of deep learning techniques [55] taking into account the semantic features of formalized statements significantly improves the premise selection procedure [38] which is at the heart of efficient ATP proof search. The semantic features preserved in the MML, in particular intermediate proof steps, can be used to select interesting lemmas for ATP proofs [42, 44]. Machine learning methods tested over the MML were also applied to define metrics between proofs [8, 41] as well as to align concepts across the libraries of different proof assistants [22]. There were also attempts to translate the MML contents to the formalisms of other proof systems [39].
Finally, it is worthwhile to mention most recent progress in parsing informal mathematics where machine learning methods refer to the correspondence of the formal MML articles and their informal English rendering automatically generated for the Formalized Mathematics journal [43].
Conclusion
To recapitulate the milestone role of establishing and developing formal libraries, the Mizar Mathematical Library in particular, let us recall a remark by F. Wiedijk, which is very accurate in the context of the MML:
We claim that the library is much more important [than] the system. A good system without a library is useless. A good library for a bad system still is very interesting (the system might be improved or the library might be ported to a different, better, system). So the library is what counts. [92]
Notes
 1.
 2.
 3.
 4.
 5.
Current MML Ver. 5.40.1289 is distributed with Mizar system Ver. 8.1.05.
 6.
The Library Committee of the Association of Mizar Users was officially established on November 11, 1989 with the main aim to collect Mizar articles and to organize them into a central repository (called Main Mizar Library at that time as multiple repositories were supposed to be created, being a direct successor of Central Archive of Mizar Texts coming with Mizar4).
 7.
A constructor is a unique number representing a given formal object with respect to its article’s environment.
 8.
In Mizar, a registration is an umbrella term for three kinds of automation techniques concerning the use of notions represented as adjectives, i.e. existential registrations, conditional registrations, as well as term adjectives registrations, also known as functorial registrations, cf. [34].
 9.
With time, the initial articles were thoroughly adjusted, 24 articles from that period were entirely removed from today’s library. Overall, the Library Committee decided on removing 52 original articles from the library.
 10.
 11.
 12.
E.g. at ISLA 2010 (http://ali.cmi.ac.in/isla2010/workshops/mizar/), CADE 2013 (http://mizar.uwb.edu.pl/CADE23tutorial/), CICM 2016 (http://mizar.org/cicm_tutorial/).
 13.
E.g. at the TYPES 2007 Summer School (http://typessummerschool07.cs.unibo.it/#mizar), ESSLLI 2012 (http://www.esslli2012.pl/index.php?id=177).
 14.
The most recent and uptodate is F. Wiedijk’s “Writing a Mizar article in nine easy steps” (https://www.cs.ru.nl/F.Wiedijk/mizar/mizman.pdf). Other manuals are available at http://mizar.uwb.edu.pl/project/bibliography.html.
References
 1.
Abad, P., Abad, J.: The hundred greatest theorems (1999). http://www.cs.ru.nl/~freek/100/
 2.
Adams, A.A., Davenport, J.H.: Copyright issues for MKM. In: Asperti et al. [7], pp. 1–16. https://doi.org/10.1007/9783540278184_1
 3.
Alama, J., Brink, K., Mamane, L., Urban, J.: Large formal wikis: issues and solutions. In: Davenport et al. [19], pp. 133–148. https://doi.org/10.1007/9783642226731_10
 4.
Alama, J., Kohlhase, M., Mamane, L., Naumowicz, A., Rudnicki, P., Urban, J.: Licensing the Mizar mathematical library. In: Davenport, J.H., Farmer, W.M., Urban, J., Rabe, F. (eds.) Proceedings of the 18th Calculemus and 10th International Conference on Intelligent Computer Mathematics. MKM’11, Lecture Notes in Computer Science, vol. 6824, pp. 149–163. Springer, Berlin, Heidelberg (2011). https://doi.org/10.1007/9783642226731_11
 5.
Alama, J., Mamane, L., Urban, J.: Dependencies in formal mathematics: applications and extraction for Coq and Mizar. In: Jeuring, J., Campbell, J.A., Carette, J., Reis, G.D., Sojka, P., Wenzel, M., Sorge, V. (eds.) Intelligent Computer Mathematics—11th International Conference, AISC 2012, 19th Symposium, Calculemus 2012, 5th International Workshop, DML 2012, 11th International Conference, MKM 2012, Systems and Projects, Held as Part of CICM 2012, Bremen, Germany, July 8–13, 2012. Proceedings, Lecture Notes in Computer Science, vol. 7362, pp. 1–16. Springer (2012). https://doi.org/10.1007/9783642313745_1
 6.
Alama, J.: Mizaritems: exploring finegrained dependencies in the Mizar mathematical library. In: Davenport et al. [19], pp. 276–277. https://doi.org/10.1007/9783642226731_19
 7.
Asperti, A., Bancerek, G., Trybulec, A. (eds.): Mathematical Knowledge Management. In: Proceedings of Third International Conference, MKM 2004, Bialowieza, Poland, September 19–21, 2004, Lecture Notes in Computer Science, vol. 3119. Springer (2004)
 8.
Aspinall, D., Kaliszyk, C.: Towards formal proof metrics. In: Stevens, P., Wasowski, A. (eds.) 19th International Conference on Fundamental Approaches to Software Engineering (FASE 2016). Lecture Notes in Computer Science, vol. 9633, pp. 325–341. Springer (2016). https://doi.org/10.1007/9783662496657_19
 9.
Bancerek, G., Byliński, C., Grabowski, A., Korniłowicz, A., Matuszewski, R., Naumowicz, A., Pąk, K., Urban, J.: Mizar: Stateoftheart and beyond. In: Kerber et al. [46], pp. 261–279. https://doi.org/10.1007/9783319206158_17
 10.
Bancerek, G., Hryniewiecki, K.: Segments of natural numbers and finite sequences. Formalized Math 1(1), 107–114 (1990). http://fm.mizar.org/19901/pdf11/finseq_1.pdf
 11.
Bancerek, G.: Information retrieval and rendering with MML Query. In: J. Borwein, W. Farmer (eds.) Mathematical Knowledge Management. Lecture Notes in Computer Science, vol. 4108, pp. 266–279. Springer, Berlin Heidelberg (2006). https://doi.org/10.1007/11812289_21
 12.
Bancerek, G.: Automatic translation in formalized mathematics. Mech. Math. Appl. 5(2), 19–31 (2006)
 13.
Bancerek, G., Rudnicki, P.: A compendium of continuous lattices in Mizar: formalizing recent mathematics. J. Autom. Reason. 29(3–4), 189–224 (2002)
 14.
Blanchette, J.C., Haslbeck, M., Matichuk, D., Nipkow, T.: Mining the archive of formal proofs. In: Kerber et al. [46], pp. 3–17. https://doi.org/10.1007/9783319206158_1
 15.
Blanchette, J.C., Kaliszyk, C., Paulson, L.C., Urban, J.: Hammering towards QED. J. Formaliz. Reason. 9(1), 101–148 (2016). https://doi.org/10.6092/issn.19725787/4593
 16.
Borak, E., Zalewska, A.: Mizar Course in Logic and Set Theory, pp. 191–204. Springer Berlin Heidelberg, Berlin, Heidelberg (2007). https://doi.org/10.1007/9783540730866_17
 17.
Borsuk, K.: On the homotopy types of some decomposition spaces. Bull. Acad. Polon. Sci. 18, 235–239 (1970)
 18.
Butler, R.W., Sjogren, J.A.: A PVS graph theory library. Technical Report (1998)
 19.
Davenport, J.H., Farmer, W.M., Urban, J., Rabe, F. (eds.): Intelligent Computer Mathematics: 18th Symposium, Calculemus 2011, and 10th International Conference, MKM 2011, Bertinoro, Italy, July 18–23, 2011. Proceedings, Lecture Notes in Computer Science, vol. 6824. Springer (2011). https://doi.org/10.1007/9783642226731
 20.
de Bruijn, N.: The mathematical language AUTOMATH, its usage, and some of its extensions. In: Laudet, M. (ed.) Proceedings of the Symposium on Automatic Demonstration, pp. 29–61. Springer LNM 125, Versailles, France (1968)
 21.
Gauthier, T., Kaliszyk, C., Urban, J.: Initial experiments with statistical conjecturing over large formal corpora. In: Kohlhase, A., Libbrecht P., Miller, B.R., Naumowicz, A., Neuper, W., Quaresma, P., Tompa, F.W., Suda, M. (eds.) Joint Proceedings of the FM4M, MathUI, and ThEdu Workshops, Doctoral Program, and Work in Progress at the Conference on Intelligent Computer Mathematics 2016 colocated with the 9th Conference on Intelligent Computer Mathematics (CICM 2016), Bialystok, Poland, July 25–29, 2016. CEUR Workshop Proceedings, vol. 1785, pp. 219–228. CEURWS.org (2016). http://ceurws.org/Vol1785/W23.pdf
 22.
Gauthier, T., Kaliszyk, C.: Aligning concepts across proof assistant libraries. J. Symbol. Comput. (to appear) (2017)
 23.
Gierz, G., Hofmann, K., Keimel, K., Lawson, J., Mislove, M., Scott, D.: A Compendium of Continuous Lattices. Springer, Berlin, Heidelberg, New York (1980)
 24.
Gierz, G., Hofmann, K., Keimel, K., Lawson, J., Mislove, M., Scott, D.: Continuous Lattices and Domains, Encyclopedia of Mathematics and its Applications, vol. 93. Cambridge University Press, Cambridge (2003)
 25.
Gonthier, G., Asperti, A., Avigad, J., Bertot, Y., Cohen, C., Garillot, F., Roux, S.L., Mahboubi, A., O’Connor, R., Biha, S.O., Pasca, I., Rideau, L., Solovyev, A., Tassi, E., Théry, L.: A machinechecked proof of the Odd Order Theorem. In: Blazy, S., PaulinMohring, C., Pichardie, D. (eds.) ITP. Lecture Notes in Computer Science, vol. 7998, pp. 163–179. Springer (2013)
 26.
Grabowski, A., Korniłowicz, A., Schwarzweller, C.: Equality in computer proofassistants. In: Ganzha, M., Maciaszek, L.A., Paprzycki, M. (eds.) Proceedings of the 2015 Federated Conference on Computer Science and Information Systems. Annals of Computer Science and Information Systems, vol. 5, pp. 45–54. IEEE (2015). https://doi.org/10.15439/2015F229
 27.
Grabowski, A., Korniłowicz, A., Schwarzweller, C.: On algebraic hierarchies in mathematical repository of Mizar. In: Ganzha, M., Maciaszek, L.A., Paprzycki, M. (eds.) Proceedings of the 2016 Federated Conference on Computer Science and Information Systems. Annals of Computer Science and Information Systems, vol. 8, pp. 363–371. IEEE (2016). https://doi.org/10.15439/2016F520
 28.
Grabowski, A., Moschner, M.: Managing heterogeneous theories within a mathematical knowledge repository. In: Asperti et al. [7], pp. 116–129. https://doi.org/10.1007/9783540278184_9
 29.
Grabowski, A., Schwarzweller, C.: On duplication in mathematical repositories. In: Autexier, S., Calmet, J., Delahaye, D., Ion, P.D.F., Rideau, L., Rioboo, R., Sexton, A.P. (eds.) Intelligent Computer Mathematics, 10th International Conference, AISC 2010, 17th Symposium, Calculemus 2010, and 9th International Conference, MKM 2010, Paris, France, July 5–10, 2010. Proceedings, Lecture Notes in Computer Science, vol. 6167, pp. 300–314. Springer (2010). https://doi.org/10.1007/9783642141287_26
 30.
Grabowski, A., Schwarzweller, C.: Revisions as an essential tool to maintain mathematical repositories. In: Proceedings of the 14th Symposium on Towards Mechanized Mathematical Assistants: 6th International Conference. Calculemus ’07 / MKM ’07, pp. 235–249. SpringerVerlag, Berlin, Heidelberg (2007). https://doi.org/10.1007/9783540730866_20
 31.
Grabowski, A.: Expressing the notion of a mathematical structure in the formal language of Mizar. In: Gruca, A., Czachorski, T., Harezlak, K., Kozielski, S., Piotrowska, A. (eds.) ManMachine Interactions 5. ICMMI 2017, vol. 659, pp. 261–270. Springer (2017). https://doi.org/10.1007/9783319677927_26
 32.
Grabowski, A.: Mechanizing complemented lattices within Mizar type system. J. Autom. Reason. 55(3), 211–221 (2015). https://doi.org/10.1007/s1081701593335
 33.
Grabowski, A.: Lattice theory for rough sets—a case study with Mizar. Fundam. Inform. 147(2–3), 223–240 (2016). https://doi.org/10.3233/FI20161406
 34.
Grabowski, A., Korniłowicz, A., Naumowicz, A.: Mizar in a nutshell. J. Formaliz. Reason. 3(2), 153–245 (2010)
 35.
Grabowski, A., Korniłowicz, A., Naumowicz, A.: Four decades of Mizar. J. Autom. Reason. 55(3), 191–198 (2015). https://doi.org/10.1007/s1081701593451
 36.
Hales, T., Adams, M., Bauer, G., Dang, T.D., Harrison, J., Hoang, L.T., Kaliszyk, C., Magron, V., McLaughlin, S., Nguyen, T., et al.: A formal proof of the Kepler conjecture. Forum of Mathematics, Pi 5 (2017). https://doi.org/10.1017/fmp.2017.1
 37.
Iancu, M., Kohlhase, M., Rabe, F., Urban, J.: The Mizar mathematical library in OMDoc: translation and applications. J. Autom. Reason. 50(2), 191–202 (2013). https://doi.org/10.1007/s1081701292714
 38.
Irving, G., Szegedy, C., Alemi, A.A., Eén, N., Chollet, F., Urban, J.: Deepmath—deep sequence models for premise selection. In: Lee, D.D., Sugiyama, M., von Luxburg, U., Guyon, I., Garnett, R. (eds.) Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems 2016, December 5–10, 2016, Barcelona, Spain, pp. 2235–2243 (2016). http://papers.nips.cc/paper/6280deepmathdeepsequencemodelsforpremiseselection
 39.
Kaliszyk, C., Pąk, K., Urban, J.: Towards a Mizar environment for Isabelle: foundations and language. In: Avigad, J., Chlipala, A. (eds.) Proceedings of the 5th ACM SIGPLAN Conference on Certified Programs and Proofs, Saint Petersburg, FL, USA, January 20–22, 2016, pp. 58–65. ACM (2016). https://doi.org/10.1145/2854065.2854070
 40.
Kaliszyk, C., Schulz, S., Urban, J., Vyskočil, J.: System description: E.T. 0.1. In: Felty, A.P., Middeldorp, A. (eds.) Proceedings of 25th International Conference on Automated Deduction (CADE’15). LNCS, vol. 9195, pp. 389–398. Springer (2015). https://doi.org/10.1007/9783319214016_27
 41.
Kaliszyk, C., Urban, J., Vyskočil, J.: Efficient semantic features for automated reasoning over large theories. In: Yang, Q., Wooldridge, M. (eds.) Proceedings of the 24th International Joint Conference on Artificial Intelligence (IJCAI’15), pp. 3084–3090. AAAI Press (2015)
 42.
Kaliszyk, C., Urban, J., Vyskočil, J.: Lemmatization for stronger reasoning in large theories. In: Lutz, C., Ranise, S. (eds.) Frontiers of Combining Systems (FroCoS’15). LNCS, vol. 9322, pp. 341–356. Springer (2015). https://doi.org/10.1007/9783319242460_21
 43.
Kaliszyk, C., Urban, J., Vyskočil, J.: System description: Statistical parsing of informalized Mizar formulas (to appear) (2017)
 44.
Kaliszyk, C., Urban, J.: Learningassisted theorem proving with millions of lemmas. J. Symbol. Comput. 69, 109–128 (2015). https://doi.org/10.1016/j.jsc.2014.09.032
 45.
Kaliszyk, C., Urban, J.: MizAR 40 for Mizar 40. J. Autom. Reason. 55(3), 245–256 (2015). https://doi.org/10.1007/s1081701593308
 46.
Kerber, M., Carette, J., Kaliszyk, C., Rabe, F., Sorge, V. (eds.): Intelligent Computer Mathematics—International Conference, CICM 2015, Washington, DC, USA, July 13–17, 2015, Proceedings, Lecture Notes in Computer Science, vol. 9150. Springer (2015). https://doi.org/10.1007/9783319206158
 47.
Klein, G., Elphinstone, K., Heiser, G., Andronick, J., Cock, D., Derrin, P., Elkaduwe, D., Engelhardt, K., Kolanski, R., Norrish, M., Sewell, T., Tuch, H., Winwood, S.: seL4: Formal verification of an OS kernel. In: Proceedings of the ACM SIGOPS 22nd Symposium on Operating Systems Principles, SOSP’09, pp. 207–220. ACM, New York, NY, USA (2009). https://doi.org/10.1145/1629575.1629596
 48.
Kohlhase, M., Johansson, M., Miller, B.R., de Moura, L., Tompa, F.W. (eds.): Intelligent Computer Mathematics. In: Proceedings of the 9th International Conference, CICM 2016, Bialystok, Poland, July 25–29, 2016, Lecture Notes in Computer Science, vol. 9791. Springer (2016). https://doi.org/10.1007/9783319425474
 49.
Korniłowicz, A.: Enhancement of Mizar texts with transitivity property of predicates. In: Kohlhase et al. [48], pp. 157–162. https://doi.org/10.1007/9783319425474_12
 50.
Korniłowicz, A.: A proof of the Jordan curve theorem via the Brouwer fixed point theorem. Mech. Math. Appl. Spe. Issue Jordan Curve Theorem 6(1), 33–40 (2007)
 51.
Korniłowicz, A.: On rewriting rules in Mizar. J. Autom. Reason. 50(2), 203–210 (2013). https://doi.org/10.1007/s1081701292616
 52.
Korniłowicz, A.: Definitional expansions in Mizar. J. Autom. Reason. 55(3), 257–268 (2015). https://doi.org/10.1007/s1081701593317
 53.
Korniłowicz, A.: Flexary connectives in Mizar. Comput. Lang. Syst. Struct. 44, 238–250 (2015). https://doi.org/10.1016/j.cl.2015.07.002
 54.
Letichevsky, A.A., Lyaletski, A.V., Morokhovets, M.K.: Glushkov’s evidence algorithm. Cybern. Syst. Anal. 49(4), 489–500 (2013). https://doi.org/10.1007/s105590139534z
 55.
Loos, S., Irving, G., Szegedy, C., Kaliszyk, C.: Deep network guided proof search. In: Eiter, T., Sands, D. (eds.) LPAR21. 21st International Conference on Logic for Programming, Artificial Intelligence and Reasoning. EPiC Series in Computing, vol. 46, pp. 85–105. EasyChair (2017)
 56.
Matuszewski, R., Rudnicki, P.: Mizar: the first 30 years. Mech. Math. Appl. Special Issue on 30 Years of Mizar 4(1), 3–24 (2005)
 57.
Milewski, R.: Fundamental theorem of algebra. Formaliz. Math. 9(3), 461–470 (2001). http://fm.mizar.org/20019/pdf93/polynom5.pdf
 58.
Milner, R.: Implementation and applications of Scott’s logic for computable functions. SIGPLAN Not. 7(1), 1–6 (1972). https://doi.org/10.1145/942578.807067
 59.
Naumowicz, A., Byliński, C.: Improving Mizar texts with properties and requirements. In: Asperti, A., Bancerek, G., Trybulec, A. (eds.) Mathematical Knowledge Management, Third International Conference, MKM 2004 Proceedings. MKM’04, Lecture Notes in Computer Science, vol. 3119, pp. 290–301 (2004). https://doi.org/10.1007/9783540278184_21
 60.
Naumowicz, A., Korniłowicz, A.: Introducing Euclidean relations to Mizar. In: Ganzha, M., Maciaszek, L.A., Paprzycki,M. (eds.) Proceedings of the 2017 Federated Conference on Computer Science and Information Systems, FedCSIS 2017, Prague, Czech Republic, September 3–6, 2017. Annals of Computer Science and Information Systems, vol. 11, pp. 245–248. IEEE (2017). https://doi.org/10.15439/2017F368
 61.
Naumowicz, A., Piliszek, R.: Accessing the Mizar library with a weakly strict Mizar parser. In: Kohlhase et al. [48], pp. 77–82. https://doi.org/10.1007/9783319425474_6
 62.
Naumowicz, A.: Enhanced processing of adjectives in Mizar. In: Grabowski, A., Naumowicz, A. (eds.) Computer Reconstruction of the Body of Mathematics. Studies in Logic, Grammar and Rhetoric, vol. 18(31), pp. 89–101. University of Białystok (2009)
 63.
Naumowicz, A.: Evaluating prospective builtin elements of computer algebra in Mizar. In: Matuszewski, R., Zalewska, A. (eds.) From Insight to Proof: Festschrift in Honour of Andrzej Trybulec. Studies in Logic, Grammar and Rhetoric, vol. 10(23), pp. 191–200. University of Białystok (2007). http://mizar.org/trybulec65/
 64.
Naumowicz, A.: SATenhanced Mizar proof checking. In: Watt et al.[91], pp. 449–452. https://doi.org/10.1007/9783319084343_37
 65.
Naumowicz, A.: Teaching how to write a proof. In: FORMED 2008: Formal Methods in Computer Science Education. Budapest, March 29, 2008, pp. 91–100 (2008)
 66.
Naumowicz, A.: Tools for MML environment analysis. In: Kerber et al. [46], pp. 348–352. https://doi.org/10.1007/9783319206158_26
 67.
Naumowicz, A.: Towards standardized Mizar environments. In: Borzemski, L., Swiatek, J., Wilimowska, Z. (eds.) Information Systems Architecture and Technology: Proceedings of 38th International Conference on Information Systems Architecture and Technology—ISAT 2017—Part II, Szklarska Poręba, Poland, September 17–19, 2017. Advances in Intelligent Systems and Computing, vol. 656, pp. 166–175. Springer (2017). https://doi.org/10.1007/9783319672298_15
 68.
Naumowicz, A.: Interfacing external CA systems for Gröbner bases computation in Mizar proof checking. Int. J. Comput. Math. 87(1), 1–11 (2010). https://doi.org/10.1080/00207160701864459
 69.
Naumowicz, A.: Automating Boolean set operations in Mizar proof checking with the aid of an external SAT solver. J. Autom. Reason. 55(3), 285–294 (2015). https://doi.org/10.1007/s1081701593326
 70.
Nowak, B., Trybulec, A.: HahnBanach theorem. Formaliz. Math. 4(1), 29–34 (1993). http://fm.mizar.org/19934/pdf41/hahnban.pdf
 71.
Pąk, K.: Automated improving of proof legibility in the Mizar system. In: Watt et al.[91], pp. 373–387. https://doi.org/10.1007/9783319084343_27
 72.
Pąk, K.: Methods of lemma extraction in natural deduction proofs. J. Autom. Reason. 50(2), 217–228 (2013). https://doi.org/10.1007/s1081701292670
 73.
Pąk, K.: Improving legibility of formal proofs based on the close reference principle is NPhard. J. Autom. Reason. 55(3), 295–306 (2015). https://doi.org/10.1007/s1081701593371
 74.
Retel, K.: The class of series—parallel graphs. Part I. Formaliz. Math. 11(1), 99–103 (2003). http://fm.mizar.org/200311/pdf111/necklace.pdf
 75.
Retel, K., Zalewska, A.: Mizar as a tool for teaching mathematics. Mech. Math. Appl. Spec. Issue 30 Years of Mizar 4(1), 35–42 (2005)
 76.
Rudnicki, P., Trybulec, A.: A collection of TeXed Mizar abstracts. University of Alberta, Techical Report (1989)
 77.
Rudnicki, P., Trybulec, A.: Abian’s fixed point theorem. Formaliz. Math. 6(3), 335–338 (1997). http://fm.mizar.org/19976/pdf63/abian.pdf
 78.
Rudnicki, P., Trybulec, A.: Mathematical knowledge management in Mizar. In: Proceedings of MKM 2001 (2001)
 79.
Sutcliffe, G.: The 8th IJCAR automated theorem proving system competition—CASCJ8. AI Commun. 29(5), 607–619 (2016). https://doi.org/10.3233/AIC160709
 80.
Tarski, A.: On wellordered subsets of any set. Fundam. Math. 32, 176–183 (1939)
 81.
The QED manifesto. In: Bundy, A. (ed.) CADE. Lecture Notes in Computer Science, vol. 814, pp. 238–251. Springer (1994)
 82.
Thomasse, S.: On betterquasiordering countable seriesparallel orders. Trans. Am. Math. Soc. 352(6), 2491–2505 (2000)
 83.
Trybulec, A.: Algebra of normal forms is a Heyting algebra. Formaliz. Math. 2(3), 393–396 (1991). http://fm.mizar.org/19912/pdf23/heyting1.pdf
 84.
Trybulec, A.: Algebra of normal forms. Formaliz. Math. 2(2), 237–242 (1991). http://fm.mizar.org/19912/pdf22/normform.pdf
 85.
Urban, J.: XMLizing Mizar: making semantic processing and presentation of MML easy. In: Kohlhase, M. (ed.) Mathematical Knowledge Management, 4th International Conference, MKM 2005, Bremen, Germany, July 15–17, 2005, Revised Selected Papers. Lecture Notes in Computer Science, vol. 3863, pp. 346–360. Springer (2005). https://doi.org/10.1007/11618027_23
 86.
Urban, J.: MPTP—motivation, implementation, first experiments. J. Autom. Reason. 33(3–4), 319–339 (2004). https://doi.org/10.1007/s1081700462451
 87.
Urban, J.: MoMM—fast interreduction and retrieval in large libraries of formalized mathematics. Int. J. Artif. Intell. Tools 15(1), 109–130 (2006). https://doi.org/10.1142/S0218213006002588
 88.
Urban, J.: MPTP 0.2: design, implementation, and initial experiments. J. Autom. Reason. 37(1–2), 21–43 (2006). https://doi.org/10.1007/s1081700690323
 89.
Urban, J., Sutcliffe, G.: ATPbased crossverification of Mizar proofs: method, systems, and first experiments. Math. Comput. Sci. 2(2), 231–251 (2008). https://doi.org/10.1007/s1178600800537
 90.
Urban, J., Rudnicki, P., Sutcliffe, G.: ATP and presentation service for Mizar formalizations. J. Autom. Reason. 50(2), 229–241 (2013). https://doi.org/10.1007/s108170129269y
 91.
Watt, S.M., Davenport, J.H., Sexton, A.P., Sojka, P., Urban, J. (eds.): Intelligent Computer Mathematics—International Conference, CICM 2014, Coimbra, Portugal, July 7–11, 2014. Proceedings. Lecture Notes in Computer Science, vol. 8543. Springer (2014). https://doi.org/10.1007/9783319084343
 92.
Wiedijk, F.: Estimating the cost of a standard library for a mathematical proof checker (2001). http://www.cs.ru.nl/~freek/notes/mathstdlib2.pdf
Acknowledgements
A. Naumowicz’s work on this paper was partially supported by the project \(\hbox {N}^\circ \) 2017/01/X/ST6/00012 financed by the National Science Center, Poland. K. Pąk’s work was supported by the resources of the Polish National Science Center granted by decision \(\hbox {N}^\circ \) DEC2015/19/D/ST6/01473. The English version of Formalized Mathematics was financed under agreement 548/PDUN/2016 with the funds from the Polish Minister of Science and Higher Education for the dissemination of science. The processing and analysis of the Mizar library has been performed using the infrastructure of the University of Bialystok High Performance Computing Center (http://uco.uwb.edu.pl).
Author information
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Bancerek, G., Byliński, C., Grabowski, A. et al. The Role of the Mizar Mathematical Library for Interactive Proof Development in Mizar. J Autom Reasoning 61, 9–32 (2018). https://doi.org/10.1007/s1081701794406
Received:
Accepted:
Published:
Issue Date:
Keywords
 Proof assistant
 Repository
 Mizar Mathematical Library